Technical Details
-----------------
-----------------


Running the Program
-------------------

1) The DATAFILE=, COVFILE=, OUTMODEL=, and OUTNULL= arguments in the
   driver file tell the program to look in the current directory for
   the files. If your data file resides in a subdirectory called
   /data, say, then you may supply this information in the DATAFILE=
   argument (DATAFILE=data/data.txt for instance). More generally
   speaking, you may supply the full path name(s) of the input/output
   files in the driver file.

2) If the driver file does not reside in the same directory within
   which you are running R, you may supply the full path name of the
   driver file when prompted after typing source("gc") or source("gcf").


Data Issues
-----------

1) The subjects must appear in the same order in both the data file
   and the covariate file. The programs will not verify that the
   subject IDs match up in both files.

2) For binary response variables (MODELTYPE=0), the variable needs to
   coded as 0's and 1's.

3) No missing data is allowed in the response variable.

4) In the data file, use the value of 0 in both columns for missing
   loci data. 

5) Factor valued covariates will be treated as continuous variables.

6) In the covariate file, use NA for missing values.

7) The covariates are normalized for use in the models, however, the
   unnormalized data is used for the summary statistics.


Models and Output
-----------------

1) For the model output file, the header contains information that was
   supplied to the program by you via the driver file. Results from
   each of the models follow the header, including test statistics,
   p-values, and Bonferroni/FDR p-value adjustments. 

2) The test statistics for all of the tests except for the "omnibus"
   test in the interaction models are the usual t-statistics from
   regression, adjusted by the square-root of the estimated inflation
   factor. For the GC program, the p-values are computed by comparing 
   the test statistics to a chi-square distribution with 1 degree of 
   freedom. For the GCF program, the comparison is made to an F 
   distribution with 1 and L degrees of freedom, where L is the number 
   of loci used in the estimation of the inflation factors.

3) The test statistic for the "omnibus" test in the interaction models
   is the usual F-statistic comparing the full model to the model
   including only the covariates. The p-values are computed by first
   adjusting the test statistics by the estimated omnibus inflation 
   factor. For the GC program, the p-values are computed by comparing 
   the adjusted statistic to an F distribution with numerator degrees 
   of freedom (df_cov - df_full), and denominator degrees of freedom 
   df_full, where df_cov is the degrees of fredom from the model 
   including only the covariates, and df_full is the degrees of
   freedom from the full model. For the GCF program, the denominator 
   degrees of freedom is changed to L, where L is the number of loci
   used in the estimation of the inflation factors.

4) The null model output file gives results (test statistics and
   p-values) from the null models used in the estimation of the
   inflation factor(s).

5) If a model can not be fit (due to singularity of the design matrix,
   say), then a warning message will be printed to the corresponding
   output file.

6) The Bonferroni and False Discovery Rate (FDR) p-value adjustments
   are only made for the main effects in the marginal models and the
   omnibus tests in the interaction models.

7) If the FDR cutoff does not apply (for instance, in cases where none
   of the tests are significant even at the ALPHA= level), "NA" will
   be printed in the header of the model output file.

8) In the GC program, if any of the estimated values for the inflation 
   factor(s) are less than 1, the program automatically sets the 
   corresponding inflation factor to 1.